A General Framework Based on Machine Learning for Algorithm Selection in Constraint Satisfaction Problems

Many of the works conducted on algorithm selection strategies—methods that choose a suitable solving method for particular problem—start from scratch since only few investigations reusable components such methods are found in literature. Additionally, researchers might unintentionally omit some implementation details when documenting strategy. This makes it difficult others to reproduce behavior obtained by an approach. To address these problems, we propose rely existing techniques Machine Learning realm speed-up generation strategies while improving modularity and reproducibility research. The proposed solution model is implemented domain-independent module executes core mechanism task. produced this work tested rapidly compared against time would take build similar approach scratch. We produce four novel selectors based constraint satisfaction problems verify our Our data suggest algorithms outperform best performing set test instances. For example, Multiclass Neural Network (MNN) Logistic Regression (MLR), powered neural network linear regression, respectively, reduced search cost (in terms consistency checks) heuristic (KAPPA), average, 49% instances considered work.